Frontend
Premium stateless chat experience and architecture visualization.
Next.js 15React 19TypeScriptTailwindCSSFramer Motion
Interactive system map
Architecturebuilt for grounded recommendations.
How HireSense AI transforms recruiter conversations into grounded SHL assessment recommendations through a stateless API, hybrid retrieval, and validated recommendation objects.
Interactive flow
From conversation to catalog-grounded decision.
Click any block to inspect purpose, inputs, outputs, technologies, and why that layer improves recommendation quality.
Technology stack
A production stack with clear separation of concerns.
Each layer owns one job: interface, API orchestration, retrieval evidence, embeddings, or deployment.
Backend
Validated API boundary and conversation orchestration layer.
FastAPIPydanticPython 3.12Structured logging
Retrieval
Hybrid evidence engine tuned for recall and deterministic ranking.
FAISSLexical searchMetadata scoringRRF
Embeddings
Offline semantic documents and vector index artifacts.
SentenceTransformersall-MiniLM-L6-v2Versioned artifacts
Deployment
Frontend and backend deploy independently while preserving API contracts.
DockerHugging Face SpacesEnvironment config
System characteristics
Built for trust, speed, and explainability.
The architecture keeps AI behavior grounded while giving recruiters a polished decision experience.
Stateless
Every /chat call carries the full conversation history.
Hybrid Retrieval
Semantic and lexical signals are combined before ranking.
Grounded Responses
Recommendation objects come from trusted catalog metadata only.
FastAPI
A clean, typed API surface for production serving.
FAISS
Fast local vector search over validated SHL assessment documents.
Docker
Portable backend deployment with explicit runtime artifacts.
Type Safety
TypeScript and Pydantic protect both sides of the contract.
Accessibility
Keyboard, focus, and reduced-motion behavior are first-class.
Performance metrics
The pipeline is optimized around bounded, explainable work.
The frontend shows the architecture in product language while the backend keeps retrieval deterministic and testable.
Response Time
< 10s
Client timeout protects the product experience.
Retrieval Strategy
Hybrid
Semantic + lexical + metadata evidence.
Vector Search
FAISS
Prebuilt index loaded by the backend.
Hybrid Ranking
RRF
Rank-based fusion across different score scales.
Pipeline
Grounded
Validated SHL recommendation objects only.
Deployment overview
Frontend, backend, and catalog artifacts deploy as clean boundaries.
The user experience can evolve independently from the FastAPI backend and SHL catalog pipeline.
Frontend
Next.js product UI
Backend
FastAPI recommendation service
SHL Catalog
Validated catalog + FAISS artifacts